Tuesday, February 10, 2026
8:00 am – 9:00 am
Presenter: Dr. Maher Alghalayini, Lawrence Berkeley National Laboratory
Abstract:
Materials synthesis processes typically involve large parameter spaces with numerous tunable process parameters, making it challenging to systematically understand how these parameters affect the quality of synthesized materials. Traditional approaches require exhaustive experimental campaigns, where parameters are varied one at a time. An approach that is time-consuming and resource-intensive. Machine learning and artificial intelligence techniques are revolutionizing this challenge through closed-loop optimization, enabling efficient exploration of complex synthesis spaces and rapid establishment of synthesis-property relationships.
This work presents two complementary demonstrations of AI-driven robotic platforms that utilize Bayesian optimization to efficiently learn the effects of process parameters on material quality. The first targets metal halide perovskites (MHPs) for next-generation solar cells, where sensitivity to processing conditions creates a challenging parameter space. Our robotic platform utilizes a multimodal data fusion metric, combining UV-Vis transmission, photoluminescence (PL) spectra, and PL imaging data, to characterize material quality. Additionally, a distribution entropy-based acquisition function selects the most informative future experiments, enabling exploration of a four-dimensional synthesis space with sampling of less than 1% of over 5,000 possible combinations.
The second application addresses atomic layer deposition (ALD) in semiconductor manufacturing, where identifying the experimental parameter space that supports optimal growth properties (the “ALD window”) traditionally requires numerous manual experiments. Our robotic plasma-enhanced ALD (PE-ALD) system incorporates Bayesian optimization algorithms to autonomously explore a twelve-dimensional parameter space for titanium oxide deposition.
Both platforms demonstrate that strategic machine learning approaches can achieve significant uncertainty reduction and accurate mapping of synthesis-property relationships with a small fraction of possible experiments. This work paves the way for rational and efficient materials discovery across diverse applications, from optoelectronic materials to nanofabrication processes, reducing reliance on exhaustive experimental campaigns and stringent environmental controls.
Bio:
Maher Alghalayini is a third-year postdoctoral researcher at the Inorganic Facility at the Molecular Foundry. His research focuses on developing and applying machine learning and artificial intelligence techniques to accelerate the discovery and characterization of materials across diverse applications. Maher began his work at Berkeley Lab on a project to predict battery lifetimes and failure distributions using domain-knowledge-informed machine learning models. Currently, Maher’s research has expanded to encompass autonomous and machine learning-guided synthesis optimization platforms. He has developed a Bayesian optimization framework for metal halide perovskite synthesis that efficiently explores complex parameter spaces using multimodal data fusion and information theory-based acquisition functions. In parallel, he has designed and implemented a closed-loop Bayesian optimization algorithm to autonomously control a robotic plasma-enhanced atomic layer deposition system, enabling the identification of optimal synthesis windows for semiconductor manufacturing.